CLASSIFICATION_SETUP.md

Product Classification Service - Setup Guide

Overview

This service uses Google Gemini Vision AI to classify product condition from uploaded images. It determines whether products are resellable, refurbishable, or scrap.

Prerequisites

• Python 3.8+
• Google Gemini API key

Installation

1. Install Dependencies

cd backend
pip install -r requirements.txt

2. Get Gemini API Key

1. Visit: https://makersuite.google.com/app/apikey
2. Sign in with your Google account
3. Click "Create API Key"
4. Copy the generated key

3. Configure Environment

# Copy the example env file
cp .env.example .env

# Edit .env and add your Gemini API key
# GEMINI_API_KEY=your_actual_api_key_here

4. Test the Service

python manage.py shell

In the Python shell:

from store.gemini_classification_service import get_classification_service

# This will test if your API key works
service = get_classification_service()
print("Service initialized successfully!")

Usage

API Endpoint

POST /api/inference/

Request:

• Content-Type: multipart/form-data
• Body: image (file)

Response:

{
  "status": "resellable"  // or "refurb" or "scrap"
}

Classification Logic

RESELLABLE

• Pristine or near-pristine condition
• No visible damage
• Can be sold as-is

REFURB (Refurbishable)

• Minor to moderate damage
• Repairable economically
• Needs cleaning or minor repairs

SCRAP

• Severe damage
• Not economically viable to repair
• Safety hazards

Frontend Integration

The Exchange.jsx page already integrates with this service:

const formData = new FormData();
formData.append('image', imageFile);

const response = await api.post('/inference/', formData, {
  headers: { 'Content-Type': 'multipart/form-data' },
});

console.log(response.data.status); // "resellable", "refurb", or "scrap"

Error Handling

The service includes robust error handling:

• If Gemini API fails, defaults to "refurb" (safe middle ground)
• Invalid responses are logged and fallback to "refurb"
• All errors are logged for monitoring

Cost Optimization

• Uses gemini-1.5-flash for cost-effective analysis
• Low temperature (0.1) for consistent results
• Max 10 tokens output (only need one word)

To use higher accuracy model, edit gemini_classification_service.py:

self.model = genai.GenerativeModel('gemini-1.5-pro')  # More accurate, higher cost

Monitoring

Check logs for classification results:

logger.info(f"Product classified as: {condition}")

Troubleshooting

"GEMINI_API_KEY not found"

• Ensure .env file exists in backend directory
• Check that GEMINI_API_KEY is set correctly
• Restart Django server after updating .env

"google-generativeai not installed"

pip install google-generativeai

Classification always returns "refurb"

• Check API key is valid
• Check internet connection
• Review logs for error messages
• Ensure image file is valid (JPEG, PNG)

Security Notes

• Never commit .env file to git (already in .gitignore)
• Never share your API key publicly
• Rotate API keys regularly
• Use environment variables in production

Production Deployment

1. Set DEBUG=False in settings
2. Use production-grade API key with rate limits
3. Add monitoring/alerting for classification failures
4. Consider adding image size limits (max 10MB recommended)
5. Add rate limiting to prevent API abuse